DanceOPD: On-Policy Generative Field Distillation

2026-06-25

AI Summary

Existing unified models for text-to-image generation and editing often suffer performance degradation due to conflicting demands across multiple capabilities. This work proposes an in-policy generative field distillation framework tailored for flow-matching models, which, for the first time, unifies diverse generative abilities as velocity fields within a shared state space. The approach enables collaborative training through sample routing and low-noise state queries based on the student model’s own trajectories. By introducing an in-policy distillation mechanism, the framework supports flexible capability composition and remains compatible with operations such as classifier-free guidance (CFG). Experiments demonstrate that the proposed method significantly enhances performance across a range of tasks—including text-to-image synthesis, local and global editing, realism-preserving field fusion, and CFG absorption—while preserving baseline generation quality.

Abstract

Modern image generation demands a single model that unifies diverse capabilities, including text-to-image (T2I), local editing, and global editing. However, these capabilities are rarely naturally aligned and often conflict. For instance, editing tends to degrade T2I performance, while global and local editing interfere with each other. Consequently, effectively composing these capabilities has become a central challenge for image generation model training. To tackle this, we introduce DanceOPD, an on-policy generative field distillation framework for flow-matching models that routes each sample to one capability field, queries one low-noise student-induced state, and trains with a simple velocity MSE objective. With each capability source defined as a velocity field over the shared flow state space, the student learns from fields queried on its own rollout states to compose expert capabilities. This formulation also absorbs operator-defined fields such as classifier-free guidance. Comprehensive experiments on T2I, editing, realism-field absorption, and CFG absorption show that our approach improves multi-capability composition, strengthening target capabilities while preserving anchor generation quality. We believe this work establishes a practical route for generative field distillation in flow-matching models.

Problem

• image generation
• multi-capability composition
• text-to-image
• local editing
• global editing

Innovation

• on-policy distillation
• generative field
• flow-matching
• multi-capability composition
• velocity field